US5548548A

US5548548A - Pass transistor for a 256 megabit dram with negatively biased substrate

Info

Publication number: US5548548A
Application number: US08/358,647
Authority: US
Inventors: Amitava Chatterjee; Jiann Liu; Purnendu Mozumder; Mark S. Rodder; Ih-Chin Chen
Original assignee: Texas Instruments Inc
Current assignee: Texas Instruments Inc
Priority date: 1994-03-21
Filing date: 1994-12-19
Publication date: 1996-08-20
Anticipated expiration: 2014-03-21
Also published as: JPH0846156A

Abstract

A design to attain a pass transistor for a 256 Mbit DRAM part. The transistor having a gate length of about 0.3 μm, a t_ox of about 85 Å, which is much thicker than the ˜65 Å t_ox for 0.25 μm logic technology, a V_WL of 3.75 V, a V_sub of -1 V, arsenic LDD and a boron concentration in the channel region of about 2.7×10¹⁷ /cm³ are the desired technological choices for 256 Mbit DRAM devices.

Description

This application Continuation of application Ser. No. 08/215,241 filed Mar. 21, 1994.

FIELD OF THE INVENTION

This invention relates generally to pass transistors used in the cells of dynamic random access memory (DRAM) devices and particularly relates to the parameters of pass transistors used in 256 Megabit DRAM devices.

DESCRIPTION OF THE RELATED ART

Leading edge design rules for process, device and manufacturing technologies make DRAMs the technology driver. Although the technological challenges are manifold: lithography, storage capacitor and isolation, to name a few, a robust pass transistor design is equally important to a successful one transistor cell DRAM. The maximum stored or charged voltage V_ch that a pass transistor can write to the capacitor determines the important storage charge of the cell, while the off-state leakage current I_off through the pass transistor, among other leakages, determines the important data retention time. Circuit techniques to reduce subthreshold leakage in DRAMs are being explored. See T. Kawahara et al, Subthreshold Current Reduction for Decoded-Driver by Self-Reverse Biasing, IEEE JSSC, Vol. 28, No. 11, November 1993, pp. 1136-43, incorporated herein by reference, but such approaches are not yet established for the pass transistor.

The design considerations for a DRAM pass transistor differ from those for a high performance transistor for logic. The design objective of maximizing the performance figure-of-merit for logic transistors, see M. Rodder et al, Oxide Thickness Dependence of Inverter Delay and Device Reliability for 0.25 u*m CMOS Technology, IEDM Tech. Dig., 1993, pp. 879-882, incorporated herein by reference, is not relevant for the DRAM pass transistor design.

Increasing or reducing the values of the design variables have gains and penalties specific to pass transistor designs. For example, increasing the substrate concentration has the desirable effect of reducing subthreshold leakage, but, on the other hand, has the undesirable effect of increasing the body-effect, which, in turn, reduces the stored charge or the voltage V_ch to Which the cell is charged. Likewise, reducing the oxide thickness has the desirable effect of reducing the rolloff in threshold voltage V_t at short gate lengths and thereby reducing I_off, but the undesirable effect is that the booted wordline voltage has to be reduced to maintain reliability of the thinner oxide. The reduced wordline voltage results in a lower V_ch.

Also, unlike logic devices, DRAMs use a substrate bias. A substrate back bias, as opposed to a zero substrate bias, can also have conflicting effects on the pass transistor design. On one hand, the subthreshold slope will improve which tends to reduce I_off ; on the other hand, the V_t rolloff gets worse. Due to these conflicting restraints, choosing the optimum conditions for oxide thickness t_ox, wordline voltage V_WL, substrate voltage V_sub and channel concentration is a non-trivial device design issue.

Further, all of the extremely large number of pass transistors (2²⁸ or 268,435,456 for a 256M DRAM part) must be within specification for the chip to be completely functional and to pass quality and reliability standards. As the gate length L is scaled, both I_off and V_ch influence a novel design optimization aimed at maximizing the probability that all the transistors are within specification despite statistical variations in the gate length L.

SUMMARY OF THE INVENTION

The present invention achieves a pass transistor for a single cell dynamic random access memory device or part that facilitates fabrication of over 256 million bits of information on a single chip of silicon semiconductor. The design variables for pass transistor design are defined by a combination of lengths, thicknesses, voltages and implanted impurity concentrations in the silicon. These parameters and their ranges distinctly describe the desired pass transistor.

The claimed invention thus encompasses a dynamic random access memory formed in a semiconductor substrate with a plurality of one transistor cells. The pass transistor has source and drain implant regions on either side of a channel region in the substrate and a gate overlying and separated from the channel region by a dielectric layer. The pass transistor comprises a patterned gate length L of about 0.3 μm, a dielectric thickness of about 85 Å, the source and drain being lightly doped with arsenic, the gate carrying a wordline voltage of about 3.75 volts, the substrate being biased at about -1 volts, and the substrate being doped with a boron concentration in the channel region of about 2.7×10¹⁷ /cm³.

The range of values for these parameters can be obtained additionally for pass transistors having a gate length L=0.25 μm by providing a boron pocket implant near the source and drain regions.

BRIEF DESCRIPTION OF THE DRAWING FIGURE

FIG. 1 is a graph of two transistor I-V curves;

FIG. 2 is a graph of time-to-breakdown, t_BD, VS 1/E_ox, the inverse of the electric field in the oxide, for t_ox =77 Å, 85 Å, 96 Å and 103 Å;

FIG. 3 is a graph of intrinsic E_ox for t_BD =10 year, operating E_ox, and V_WL VS t_ox ;

FIG. 4 is a graph of drain current vs drain voltage indicating gate-induced drain leakage GIDL at 300K for both arsenic As lightly doped drains LDD and phosphorous P LDD at t_ox =77 Å, 85 Å and 96 Å;

FIG. 5 is a graph of I_off VS V_Tlin for phosphorous and arsenic LDD at t_ox =96 Å;

FIG. 6 is a graph of gate length L vs V_T -dose with vertical arrows marking L_margin at V_T -doses of 3×10¹² and 4.5×10¹² for a target L=0.25 μm.

FIG. 7 is a graph of L_margin VS V_T-dose for a target L=0.3 μm. Here t_ox =85 Å and V_sub =-1 V;

FIG. 8 is a graph of peak L_margin vs t_ox for L=0.3 μm and 0.25 μm. The lower L=0.3 μm line is when V_ch >0.85 V_cc. The bottom curve plots D_W vs t_ox for L=0.3 μm;

FIG. 9 is a graph of peak L_margin and D_W vs V_sub for L=0.3 μm and t_ox of 85 Å and 96 Å, with the top curves plotted to L_margin and the bottom two curves plotted to D_W ;

FIG. 10 is a graph of L_margin VS target L for a conventional As LDD (experimental) and pocket implant (simulation). V_sub =-1 V;

FIG. 11 is a graph of I_off vs V_Tlin for two conventional transistors and one pocket implant transistor t_ox =96 Å;

FIG. 12 is a graph of V_Tlin -V_sat vs gate length for a shallow S/D (N⁺ masked away from gate) and a deep S/D (n⁺ after sidewall oxide);

FIG. 13 is a schematic representation of a one transistor storage cell used in a DRAM memory device;

FIG. 14 is a cross-sectional view of an intermediate stage of fabricating a pass transistor used in the storage cell of FIG. 13 depicting a patterned gate formed over gate oxide on a substrate;

FIG. 15 is a cross-sectional view of a completed pass transistor used in the storage cell of FIG. 13 formed using one process and used to obtain the experimental data presented in this patent;

FIG. 16 is a cross-sectional view of another embodiment of a pass transistor used in the storage cell of FIG. 13 formed in a more conventional manufacturing process;

FIG. 17 is a cross-sectional view of a pass transistor represented in FIG. 13 in a plane normal to the cross-sections depicted in FIGS. 14-16.

DETAILED DESCRIPTION

Sample pass transistors were fabricated on (100) oriented p-type wafers with 10-15 Ω-cm. After standard well formation and LOCOS isolation, threshold voltage V_T adjust implant with boron B, 20 Kev, with V_T -dose ranging from 3×10 ¹² cm^-2 to 8×10¹² cm ^-2 was done. Gate oxides ranging from 77 Å to 103 Å were grown. LDD implants were As or P, 4×10¹³ cm^-2. Most of the wafers are with LDD S/D implant only. The S/D anneal occurs at 900° C. for 20 minutes, followed by metalization and sintering in forming gas at 450° C for 30 minutes.

In FIG. 1, typical leakage characteristics occur along curve 100 for a gate length L=0.2 μm, V_D =2.5 V and V_sub =-1 V, and typical V_T characteristics occur along curve 110 for a gate length L=0.3 μm, V_D =0.1 V and V_sub =-3 V. Both leakage specifications at L=0.2 μm and V_T specifications at L=0.3 μm are met.

Among the pass transistor design constraints, subthreshold leakage (I_off <3×10^-14 A/ μm) and charging voltage (V_ch >0.8 V_cc) are typically the most stringent requirements. Since the storage node voltage is effectively an additional substrate bias to the pass transistor, V_ch is equal to V_WL -V_Tsp, where V_Tsp is the V_T at substrate bias of V_sub -V_ch.

FIG. 2 presents the results of time-dependent dielectric-breakdown TDDB tests done to locate E_ox (intrinsic) for 10-year time-to-breakdown; this is used to obtain the applicable wordline voltage V_WL for various oxide thicknesses t_ox. The intrinsic E_ox occurs at dashed line 200, which represents 10 years in time.

FIG. 3 illustrates that the wordline voltage VwL is then selected to be the smaller of: t_ox times operation E_ox, which is 4 MV/cm less than that of the 10-year lifetime, or 1.5 times V_cc =3.75 V which is the practical limit for the booting circuit. Other pass transistor design constraints, that is to say, source-drain punch-through voltage (BVDSS) and gate-induced drain leakage (GIDL), turn out to be less stringent. The measured BVDSS at L=0.2 μm are >5 V for all devices that satisfy the I_off constraint.

FIG. 4 depicts that the gate-induced drain leakage GIDL depends on to and LDD species: As or P.

FIG. 5 depicts that, despite the higher GIDL currents indicated in FIG. 4, an arsenic LDD provides a better tradeoff between short-channel and body effects than phosphorous LDD. Arsenic and phosphorous are N-type dopants. In FIG. 5, I_off at L=0.25 μm for arsenic is clearly lower than phosphorous at a given V_Tsp at L=0.35 μm. Thus we choose arsenic as the LDD species.

Optimizing t_ox, V_sub, and V_T -dose for a target L occurs as follows. For each pair of t_ox and V_sub, the values for V_ch and I_off are determined on a dense grid of V_T -dose and L by interpolation and extrapolation from experimental data with the aid of tuned device simulations. The constraints on I_off and V_ch define an "acceptable" region.

In FIG. 6, the area between the two

curves

600 and 610 defines an example of such an acceptable region for t_ox =85 Å and V_sub =-1 V. The upper bound of the region is determined by the constraints on V_ch, while the lower bound is due to the constraints on I_off. Let us define a primary metric L_margin of a given design to be the margin in gate length L (vertical distance in FIG. 6) from the

nearest boundary curve

600 or 610. The yield of a given design is higher if L_margin is larger.

FIG. 7 shows the L_margin vs. V_T -dose corresponding to FIG. 6. To ensure that all the 2²⁸ transistors in a 256 Mbit DRAM are within specification, L_margin needs to be >0.08 μm. Arrow 700 indicates the range in V_t -dose for which L_margin ≧0.08 μm. A secondary metric, dose window D_W, defined as the range of V_T -dose at which L_margin ≧0.08 μm in FIG. 7, is a measure of the sensitivity of the DRAM yield to variations in V_T -dose.

FIG. 8 plots L_margin and D_W for t_ox ranging from 77 Å to 96 Å.

Curves

800, 810 and 820 are plotted to the L_margin axis and curve 830 is plotted to the dose window axis.

Curves

800 and 810 are for L=0.3 μm and curve 820 is for L=0.25 μm. Clearly L_margin is smaller at 96 Å for the

curves

800, 801 and 820 because V_WL is limited to 1.5×V_cc . Although at the 77 Å to 85 Å range L_margin for

curves

800 and 820 is relatively flat, curve 830 for D_W clearly shows that 85 Å is better than 77 Å. Curve 810, obtained using V_ch >0.85×V_cc, also demonstrates that 85 Å is optimum.

FIG. 9 plots a similar evaluation for V_sub. Curves 900 and 910 plot to the L_margin axis and curves 920 and 930 plot to the D_W axis. Again, for the 85 Å case, L_margin is relatively constant, but curve 920 clearly indicates that V_sub =1 V is an optimum choice.

This optimum choice depends on the specific constraints used here, such as V_ch >0.8 V_cc, booting limit of 1.5 V_cc, V_cc =2.5 V, operating E_ox of 4 MV/cm less than that for 10 year lifetime, I_off <3×10^-14 A/cm and L_margin ≧0.08 μm.

While these are reasonable constraints, the constraints may be relaxed or tightened depending on the requirements and capability of future technology. For example, if stepper technology improves then L_margin ≧0.08 μm may be relaxed to say L_margin 0.05 μm. Likewise, if sense amplifier sensitivity improves then a lower V_ch may be tolerated. However, we have delineated by example a design methodology that can be applied to the new constraints to arrive at an optimum choice consistent with the new constraints.

As for scaled design at L=0.25 μm, an immediate limitation is that L_margin <0.08 μm, see FIG. 7. Our simulation studies show that L_margin can be improved by using a pocket implant, as indicated by the pocket curve 1000 in FIG. 10.

Additional pass transistor samples were prepared with the previously stated values and a pocket implant of boron implanted at an angle of zero degrees. Boron is a P-type dopant.

In FIG. 11, initial experimental results indicate that a pocket counter dopant implant of boron at 0 degrees angle in a pass transistor with regular S/D junctions performs better than the transistors with only regular S/D junctions. These data support the conclusion that boron pocket implants with a shallow S/D junction should provide the large design margin indicated in FIG. 10. Using the boron counter dopant or pocket implant may be necessary to obtain a useful product when scaling the target gate length from 0.3 μm to 0.25 μm. The pocket implant will eliminate or reduce the boron V_t adjust implant dose.

The graph of FIG. 12 demonstrates that a shallow S/D junction depth of X_j =600 Å reduces the drain-induced barrier lowering V_Tlin -V_Tsat compared to a deep S/D junction.

In FIG. 13, pass transistor 1300 has a source/drain 1302 connected to bitline 1304 and a source/drain 1306 connected to storage node 1308. Gate 1310 of transistor 1300 connects to wordline 1312 and channel 1314 of transistor 1300 connects to or is formed in substrate 1316. Storage node 1308 is formed on one side of capacitor 1318 with the other side of the capacitor being formed by a field plate 1320.

In FIG. 14, transistor 1400 is in an intermediate step or stage of fabrication. The substrate P-type material 1402 carries a gate oxide 1404 having a thickness of t_ox . The gate oxide 1404 in turn carries a polysilicon gate material 1406. The gate material 1406 extends over the array of memory cells to form one of the wordlines. The gate 1404 has a dimension L resulting from patterning a layer of polysilicon material formed over the gate oxide. The dimension L is the gate length referred to in this patent. This stage of processing occurs before formation of an oxide over the gate, which consumes some of the polysilicon gate material and reduces the polysilicon gate length by about 10 percent.

In FIG. 15, a completed transistor 1500 represents the transistor used to obtain the experimental data in this patent. Transistor 1500 comprises N^- -type source and

drain regions

1502 and 1504 and channel region 1506 formed in P-type substrate 1508. Gate 1510 and gate oxide 1512 extend over the channel region 1506 and overlap the source and

drain regions

1502 and 1504. Gate 1510 overlaps the source and drain regions by about 200 Å optimally. During processing and after patterning of the gate 1510 on oxide layer 1512, a second oxide layer 1514 is formed over the gate. The gate oxide 1512 and the second oxide 1514 are patterned to their final depicted shape, and a third oxide 1516 is formed over the existing structures. Then an arsenic or phosphorus implant is done to obtain the source and

drain regions

1502 and 1504.

To obtain the pocket implant experimental data represented in FIGS. 11 and 12, some of the sample transistors were processed differently before application of the full thickness of oxide layer 1514. The gate oxide 1512, the polysilicon gate 1510 and a part of oxide 1514 were formed on the substrate 1508. A pocket implant of boron was performed to effect halos or

pockets

1520 and 1522, represented by dashed lines, near the respective source and

drain regions

1502 and 1504. This implant occurred at an implant angle of zero degrees; other implant angles could be used as desired.

In FIG. 16, transistor 1600 represents a more conventional structure. Source and

drain regions

1602 and 1604, formed of N^- -type material, and channel 1606 are formed in P-type substrate material 1608. Gate oxide 1610 overlays the substrate 1608 and carries gate 1612. Gate 1612 extends over at least part of the source and

drain regions

1602 and 1604. A layer of oxide 1614 overlays all the existing structures.

In FIG. 17, a polysilicon wordline 1700 extends over oxide 1702 and substrate 1704. Dimension 1706 indicates a dimension of the gates of

transistors

1400, 1500 and 1600 normal to the gate length L depicted in elevation in FIGS. 14, 15 and 16. The experimental data described in this application came from transistors having a dimension 1706 of about 10 μm. This dimension will be about 0.3 μm in the DRAM array. As a result of this, some of the transistor characteristics may be slightly different causing the optimum design to shift accordingly.

The

transistors

1400, 1500 and 1600 all can be formed with the parameters of channel length L=0.3 μm, oxide thickness t_ox =85 Å, wordline voltage V_WL =3.75 volts, substrate bias voltage V_sub =-1 volt, arsenic LDD and a substrate doped with boron in the channel region to a concentration of about 2.7×10¹⁷ /cm³ as the optimum technological choices to obtain a workable transistor for the 256 Megabit DRAM.

The desired parameters and their ranges occur as follows: Gate length L: 0.3 μm±0.08 μm Gate oxide thickness: 85 Å±5 Å Boron V_t adjust: 3.5×10¹² /cm² ±2×10¹² /cm², 20 Kev±2 Kev* Arsenic: 4.0×10¹³ /cm² ±1×1¹³ /cm², 30 Kev±2 Kev*

* Implant energy is dependent upon the screen oxide thickness Junction depth: ≈0.06 μm±0.02 μm Word line voltage V_WL : 3.75 Volts 0.2 Volts Substrate voltage V_sub : -1 Volt±0.5 Volt Boron concentration in the channel region: 1.5×10¹⁷ /cm³ to 5×10¹⁷ /cm³ with a target of 2.7×10¹⁷ /cm³.

If the gate length L is scaled to 0.25 μm±0.05 μm then the pass transistor can be designed with a boron pocket implant of 10¹³ /cm² ±2×10¹² /cm² and with all of the remaining above parameters, except that the V_t -adjust dose may be reduced to 1.5×10¹² /cm² ±0.5×10¹² /cm² or eliminated and the substrate voltage V_sub may be more negative: -1.5 volt±0.5 volt.

Claims

We claim:

1. A dynamic random access memory device formed in a substrate with a one transistor cell pass transistor having source and drain implant regions on either side of a channel region in the substrate and a gate overlying and separated from the channel region by a dielectric, the pass transistor comprising a patterned gate length L of about 0.3 μm, a dielectric thickness of about 85 Å, the source and drain being lightly doped with arsenic, the gate carrying a wordline voltage of about 3.75 volts, the substrate being biased at about -1 volts and a boron concentration in the channel region of about 2.7×10¹⁷ /cm³.

2. A dynamic random access memory device formed in a substrate with a one transistor cell pass transistor having source and drain implant regions on either side of a channel region in the substrate and a gate overlying and separated from the channel region by a dielectric, the pass transistor comprising a patterned gate length L of 0.3 μm±0.08 μm, a gate dielectric thickness of 85 Å±5 Å, the source and drain being lightly doped with arsenic at a dose of 4.0×10¹³ /cm² ±1×10¹³ /cm², the gate carrying a wordline voltage of 3.75 volts ±0.2 volt, the source and drain regions having a depth in the substrate of 0.06 μm±0.02 μm, the substrate containing boron in the channel region at a concentration of from 1.5×10¹⁷ /cm³ to 5×10¹⁷ /cm³ and the substrate being biased at -1 volt±0.5 volt.

3. A dynamic random access memory device formed in a substrate with a one transistor cell pass transistor having source and drain implant regions on either side of a channel region in the substrate and a gate overlying and separated from the channel region by a dielectric, the pass transistor comprising a patterned gate length L of about 0.25 μm, a dielectric thickness of about 85 Å, the source and drain being lightly doped with arsenic, the gate carrying a wordline voltage of about 3.75 volts, there being a boron pocket implant near the source and drain occurring at an implant angle of a certain degree, there being a boron concentration in the channel region of about 10¹⁷ /cm³ not counting the contribution from the boron pocket implant and the substrate being biased at about -1.5 volts.

4. The device of claim 3 in which the implant angle is zero degrees.

5. The device of claim 3 in which the gate length L is 0.25 μm±0.05 μm.

6. The device of claim 3 in which the boron pocket implant is 10¹³ /cm² ±2×10¹² /cm².

7. The device of claim 3 in which there is no boron V_t -adjust dose.

8. A dynamic random access memory device formed in a substrate with a one transistor cell pass transistor having source and drain implant regions on either side of a channel region in the substrate and a gate overlying and separated from the channel region by a dielectric, the pass transistor comprising a patterned gate length L of about 0.3 μm, a dielectric thickness of about 85 Å, the source and drain being lightly doped with an N-type dopant, the gate carrying a wordline voltage greater than the supply voltage, the substrate being biased at a negative voltage and a P-type dopant concentration in the channel region of about 1.5×10¹⁷ /cm³ to 5×10¹⁷ /cm³.

9. The device of claim 8 in which the N-type dopant is arsenic.

10. The device of claim 8 in which the wordline voltage is about 1.5 times the supply voltage.

11. The device of claim 8 in which the substrate is biased at about -1 volts.

12. The device of claim 8 in which the P-type dopant is boron with a concentration in the channel region of about 2.7×10¹⁷ /cm³.

13. The device of claim 8 in which V_D is about 2.5 volts.